KLASIFIKASI TINGKAT PENDIDIKAN PROVINSI DI INDONESIA MENGGUNAKAN SVM DENGAN HYPERPARAMETER OPTIMIZATION OPTUNA

Abstract

Educational attainment disparity among provinces in Indonesia remains a major problem, reflected in the wide variation of Mean Years of Schooling (MYS) values. This study aims to classify the education level of 34 provinces in Indonesia into three categories (Low, Medium, High) based on socio-economic indicators for the 2016-2025 period. The method used is Support Vector Machine (SVM) with hyperparameter optimization using the Optuna framework and validated using Time Series Cross-Validation to avoid data leakage in time series data. This study compares four SVM kernel functions: RBF, Polynomial, Linear, and Sigmoid. The results show that the RBF kernel achieved the highest cross-validation accuracy of 89.23% with optimal parameters C = 999.05 and gamma = 0.00471. On the 2023-2025 testing data, the RBF kernel achieved an accuracy of 85.29% (29 out of 34 provinces correctly predicted), with macro precision of 88.7% and macro F1-score of 82.9%. Classification errors occurred in 5 provinces with MYS values around the class boundaries (9.00-9.38 years). There are significant performance differences among the four SVM kernels, with the RBF kernel proven to be the most superior in capturing the non-linear patterns of socio-economic data of Indonesian provinces.